zhongxiayan/alphazero_gomoku

Multiprocess implementation of Alphazero applied to the board game Gomoku (五子棋; 5 in a row; tic-tac-toe but with 5 pieces on a 20x20 board)

AlphaZero Gomoku

Multiprocess implementation of AlphaZero applied to the board game Gomoku (五子棋; 5 in a row; tic-tac-toe but with 5 pieces on a 20x20 board). This project was done as a final project for the 6.882 Embodied Agents class at MIT.

Alphazero

alphazero directory contains code for training, code for simulating games between trained agents, and code to interface with Piskvork.

Piskvork

piskvork_remote directory contains more code to interface with Piskvork.

Note that this is an optional interface, it may be easier to write your own CLI visualizer (especially if you don't have a Windows computer).

Piskvork is a Windows-based GUI client for interacting with Gomoku agents and is used by Gomocup AI bots; basically it has a game board that you can play on and also watch AI agents play each other.

Since Piskvork is Windows-based, I wrote piskvork_remote/pisqpipe.py, which compiles to piskvork_remote/dist/pbrain-port.exe, to relay commands to / from a Linux machine. On the Linux side, I wrote piskvork_remote/remote_brain.py to listen for the messages from piskpipe.py; piskvork/remote_brain.py is overriden by alphazero/alphazero_brain.py so that trained models can be used to respond to moves.

If you want to work with Piskvork, the documentations may be helpful.